Network industry ready to change

"Editor's note" Network after years of development, has come to the "post-Cisco" era, software definition network, network virtualization and other nouns have come to everyone, this article focuses on the traditional network architecture and SDN network of similarities and differences, the development of network virtualization, you can see, SDN manufacturers are gearing up, because of " We usually overestimate the changes in the next two years, and underestimate the changes that will happen in the next decade. "The traditional enterprise needs to face the change of the future network operation form actively, the author discusses the future direction of the network industry from two aspects of network transformation and technology transformation."

The following is the original text:

Network industry ready to change

Currently, Ethernet has played a pivotal role in every data center, and it connects each company to the Internet. Every major change in the Web world takes a lot of time, and while the public's acceptance of standards and new technologies is improving, the spread of new technologies takes more than 10 years. In the development process, enterprises need to cope with the growing demand for budget reduction, enterprise employees need automated IT infrastructure to simplify operational processes, and the voice of the network World focuses on how to migrate from traditional Ethernet to software to network and network virtualization. Now that the Internet world has been ruled by Cisco for years, the question is: How powerful are these new technologies for traditional networks? Does the customer have enough willingness to replace the vendors to adopt these new technologies?

Ethernet Fabrics (ether Optical network), as a substitute for spanning Tree Protocol architecture Network, has extensible, high-bandwidth architecture, and SDN is not so much a product as an architecture. Juniper and Arista have provided some programmable features in their product lines over the years. First, let's take a look at how SDN differs from the traditional network architecture (the next two sections explain the specific differences through the diagrams and descriptions of Joe Onisick).

Traditional network architecture

In the traditional network architecture diagram, the most important point is the separation of the control layer and the data layer. Each layer has a different task, the layer and layer together to provide the entire data forwarding, routing capabilities. Here, the control layer is responsible for device configuration and Data flow programmatic selection. When you manage a switch, you are actually dealing with the control layer of the switch. Like routing tables, spanning tree protocols, these things are calculated by the control layer. These tables are built from the message frames such as BPDU (Network Bridge Protocol data unit, exchange information between switches running STP), hello packets, and so on, according to these message frames, the switch determines the available forwarding path. Once the forwarding path of these packets is determined, the path information is sent to the data tier and is usually stored on top of the hardware. The data level usually uses the latest path information transmitted by the control level for message forwarding. This pattern is traditionally very efficient, the hardware decision-making process is very fast, the overall latency controllable and the control level can handle heavy configuration requirements.

There is no problem with this approach, and we are concerned with scalability. To prove scalability, let's take quality of service (QoS) as an example. QoS allows specific data frames to be forwarded preferentially based on the characteristics of the frame and according to the scheduling requirements. This reduces the data transfer latency that a particular traffic brings to the network congestion in some way. For example, latency-sensitive voice and video traffic is usually prioritized for high priority forwarding to ensure user experience. Traffic priority is usually determined based on the service level (CoS) or differentiated Service code point (DSCP) label in the data frame. These frames must be uniformly determined when the data frame enters the network, and the corresponding rules must be set consistently across the network, which can become unwieldy in traditional multilayer switched networks because each device requires the same configuration information.

To illustrate the current challenges of network management, we think that administrators need to be configured individually on each port of each device node in the network, which is time-consuming, error-prone, and clumsy.

In addition, there are challenges in networks where data is properly categorized and routed. For example, now we have two completely different data traffic, one is iSCSI traffic, the other is voice traffic. iSCSI as a storage traffic, usually the packet is full-size, and sometimes even large data frames, and voice traffic is usually a small packet transmission. In addition, the two types of traffic have different transmission requirements: voice traffic is very sensitive to latency, in order to ensure the quality of voice communication, while iSCSI is less sensitive to latency, but requires more bandwidth. There are almost no tools in traditional networks that can differentiate between these two types of traffic and choose different data paths to meet the specific needs of both types of traffic depending on the flow type. These problems are SDN hope to solve.

Three elements in SDN the ability to manage data frames, packet forwarding, and application strategies dynamic application of policies and management of network capabilities programmable capability

NOTE: To achieve SDN, the network architecture does not need to be open, standardized, and interactive with other networks. A suitable SDN can only meet definitions and requirements.

A SDN architecture must have the ability to manipulate data frames or packet traffic on a large scale. Typically, the hardware in SDN is typically designed to converge (the device is capable of hosting all data types, including some desired storage traffic), large, low-latency data transmission pipelines, usually called fabric. The SDN architecture itself will provide a centralized management of network and network streams from the perspective of the network layer.

This architecture is achieved by separating control from the data-tier device and providing a programmable interface for the control layer. The data-tier device accepts forwarding rules from the control layer and applies the rules to the ASIC hardware. These ASIC can be traditional ASIC or can be customized according to functional and performance requirements of the forwarding chip. The following illustration illustrates the relationships between these elements:

In the diagram, the SDN controller provides the control layer function, and the data layer is provided by the physical Exchange device. These devices can be new or can be existing devices that mount specific firmware, depending on the specific vendor and the physical schema model. One of the main advantages of architecture in the diagram is that the global visibility is greater for the control level. Except that each data-tier device relies on other neighbor devices to obtain a network topology, a single control layer will have a full network connection state. This architecture provides a full network of routing, security, and QoS configuration, but also requires programmable features. Another major feature is centralized control, which makes it easier for centralized controllers to obtain real-time application data streams, and can focus on data making decisions.

The figure only shows a small part of the SDN architecture, which focuses on the physical architecture and server location. In addition, this architecture consolidates the virtual machine environment into the SDN network. In this way, both physical and virtual devices can be unified for control policy management, which can be achieved through virtual Ethernet bridges that can be managed by SDN controllers in virtual Machine Manager. The following figure illustrates the logical architecture of this solution.

The solution of integration of virtual network system and physical network system is clearly described in the figure. This architecture is extremely important because both the physical data layer and the virtual data layer are managed uniformly by the control layer, and when a virtual machine migrates, he can ignore the underlying network architecture. This feature is important for a policy virtualization environment because, if finer granularity of control is achieved, control policies can be placed on the virtual machine port and take effect until the virtual machine is out of the net.

Note: The above two graphs are all conceptual maps of the SDN architecture. In addition to the individual control node solutions, there are many other architectures, described in this article are the most common applications.

In the architecture, the centralized control and programmable interface in the control layer can be replaced by more complicated intelligent control process to meet the complex demands of traffic optimization, security and maintenance. Traffic can be forwarded independently of each other according to network changes.

The Rise of SDN manufacturers

As many people say, with Microsoft falling, the "post-PC" era is coming. Then with the arrival of the "post-Cisco" ERA, SDN will prop up a day.

cisco--recently made a major strategic decision on its own open network environment-see "Mike Fratto talking about computer networks"

BROADCOM--SDN also amplifies the competition between custom vendors and traditional chip makers. Cisco, of course, mainly uses proprietary chips, while other Ethernet switch vendors prefer to apply Broadcom or other vendors ' chips. At the recent Dell Storage Forum meeting, the Broadcom architecture and network Cto,nick Ilyadis mainly discussed the status of network software (including SDN) in the current IT system architecture. Read the details.

brocade--has released its own blueprint for development in SDN, including support for OpenFlow and Ethernet fabrics networks.

hp--announced support for OPENFLOW,HP network Cto,saar Gillar discussed SDN's ecosystem and revealed that HP was planning to develop OpenFlow controllers and read details.

dell--is expanding its competitiveness in the online marketplace through acquisitions. Dell Network GM Dario Zamarian explains how Dell can connect with SDN and read the details.

Juniper Network--Release the latest SDN plan

Arista Network--Release the latest SDN plan

IBM publishes SDN White paper, focusing on enhancing IBM network competitiveness

Network Virtualization

In the process of discussing the future form of the network, most of the vendors put the end point in the Sdn. The agreement will no longer be what the user really needs; the improvement of the enterprise IT needs requires the change of technology, training and organizational structure. Wikibon the answer from Nicira: Nicira is often seen as a SDN start-up, with its founders playing an important role in the development of OpenFlow, but the Nicira solution focuses on the control level, It is more like server virtualization than network virtualization, which provides a layer of virtualization between applications and physical layers. With VMware's vmotion, Microsoft's dynamic migration technology is proposed, these features also allow the creation of virtual machine pools can be dynamically deployed and migrated. Network virtualization enables this flexibility to be maintained across a variety of virtual machine management platforms and can migrate any virtual machines to anywhere. This feature will have a huge impact on traditional network vendors, not just technology will be out of the hardware, it is more able to take full advantage of the new physical architecture (like Blade server cluster). The three goals of Nicira network virtualization are, to some extent, complementary to SDN:

1. Software and hardware decoupling, two or three-tier data forwarding will be separated by 2. Redefine physics with Virtual 3. Network automation allows flexibility of operation

However, most SDN solutions are still in the research and development phase, and Nicira's 24 clients include large networks, telecommunications and cloud service providers (At&t, EBay, Fidelity, NTT and Rackspace). Nicira's solution is primarily for large-scale network deployment environments, which is also the biggest difference from the OpenFlow application scenario. Nicira's Dvni (Distributed Virtualization Network architecture) builds a network virtualization platform (NVP) that decoupling virtual machines from the network through Open VSwitch. Network virtualization technology also applies tunneling protocols such as Vxlan,nvgre or STT (Stateless Transport Tunneling Protocol, Nicira being used, in the IETF drafting phase). When the SDN industry is discussing the network transformation, Nicira's perspective has been to the super large-scale next generation network system direction.

Preparing for a future network

While SDN or network virtualization has not yet reached the desired heights, corporate CIOs need to put these changes into the blueprint for the future of enterprise development. In the next few years, these technologies will help enterprises to transform smoothly from the perspective of business model. So it is never too early for it practitioners to embrace these technologies, which will enable them to transform themselves instead of being eliminated when these technologies arrive. For network management personnel, can through the "engineer will drag network transformation?" This article is about the network transformation from the perspective of operational dimension. As Bill Gates understands technological change: we tend to overestimate the changes that have taken place over the next two years, and underestimate the changes that will happen in the next decade. ”

Original link: Network revolution: Software definition Network and network virtualization (Zebian/Wei)

Free Subscription "CSDN cloud Computing (left) and csdn large data (right)" micro-letter public number, real-time grasp of first-hand cloud news, to understand the latest big data progress!

CSDN publishes related cloud computing information, such as virtualization, Docker, OpenStack, Cloudstack, and data centers, sharing Hadoop, Spark, Nosql/newsql, HBase, Impala, memory calculations, stream computing, Machine learning and intelligent algorithms and other related large data views, providing cloud computing and large data technology, platform, practice and industry information services.